On 11/06/2013 11:18 AM, Gabriel Risterucci wrote:
> 2013/11/6 Kracked_P_P---webmaster <[email protected]>
>
>> To be honest, number crunching on a GPU is better than a CPU, since GPUs
>> were designed to number crunch.  Look at the newer CUDA and ATI GPU
>> cards.  They have 32, 64, 128, 512, or more "cores" or "streams".  There
>> is a movement to create systems that are GPU based instead of CPU
>> based.  The gaming systems are almost all GPU based, since they do not
>> need to run traditional packages that a "home" or business computer
>> needs to work with.
>>
>> Well, the Excel vs. Calc speed comparison on the same system [32-bit] is
>> a different "thing" than making a 64-bit version of LO or a GPU based LO
>> package.  The difference between Calc and Excel may be the efficiency of
>> the coding.  There are still a lot of old legacy code in LO that is
>> being worked on to make it work better and much more efficient.  Just
>> saying we need to create a 64-bit version of LO to fix the "speed
>> issues" is not really solving that issue.
>>
>> As for making LO work with a GPU card, well I would not be surprised
>> that not too long from now, both Windows and Linux will have a version
>> that is GPU based.  That is one of the things that would make our
>> current systems faster without replacing the motherboard or CPU.  Just
>> buy a newer, faster, GPU for the system.  This is what the gamers do
>> currently.  The price of these faster GPUs are going down.  For $100, I
>> could buy a GPU 3 to 4 time faster than one I could buy 2 or 3 years
>> ago.  The GPU speeds per price is a much better "ratio" than the CPU
>> speed per price.  You just get more speed or number crunching power for
>> you money with a GPU card, compared to the CPU/motherboard costs.
>>
>
> ​Whoa, you're leaping way too fast here. GPU are better at number
> crunching, but only at that, and litteraly at number crunching. Branching,
> working with the rest of the hardware, interrupts, and even some kind of
> float or integer operation are not exactly their shining part.
>
> I agree that moving the *relevant* operations to GPU-based hardware can
> (and actually do) provide a very large boost, but the kind of operation
> that benefit from them is really limited. There is a reason we still have
> those general-purpose CPU around you know. Talking about windows or linux
> systems that are GPU based is a bit of a stretch.
> Even for LO Calc, formulas that are more complicated than "add x and y",
> for example using lookup functions, statistical functions and other funny
> stuff would probably not run better on GPU. One of the key point of the GPU
> (or GPGPU, general purpose GPU) performance boost is that you can handle
> *arithmetic operations* that are *massively* parallel, not that you have a
> lot of raw computing power.
>
>

The real thing needed is to make the LO "Calc coding" more efficient
than it is now.  Just giving it a new place to do the number crunching
will not solve the "true issue" of why Calc is slower than Excel. 

Once the developers clean up the legacy coding in the Calc "code
objects", and make as much as possible more efficient, then people
should be happier about how long it takes Calc to do it work.

To be strictly honest, the future will be a system that takes the best
of the CPU and GPU computing options and blend them into a system where
the OS can use the most efficient route to do the work, and then write
software packages to do the same.

Also, not too many OSs and packages can take advantage of a multi-core
system to bring all 2, 3, 4, 6, 8, or more cores, to do the processing
of the package running.  There are a few though.  I have 4 cores and I
see many times where one core is running at 90+% while the others are
running under 15 or 20% for the other packages I have open in the
background.  As I type this, each core is running between 0% and 4 % at
1.15 GHz.  So I am not taxing the CPu so I am not running at the 2.30
GHz max speed of the processor.





-- 
To unsubscribe e-mail to: [email protected]
Problems? http://www.libreoffice.org/get-help/mailing-lists/how-to-unsubscribe/
Posting guidelines + more: http://wiki.documentfoundation.org/Netiquette
List archive: http://listarchives.libreoffice.org/global/users/
All messages sent to this list will be publicly archived and cannot be deleted

Reply via email to